Featured Image Source: Kona Space Systems
Overview
In the OODA Almanac 2022 – Exponential Disruption, OODA CEO Matt Devost provided our initial position (or working hypothesis) for the exploration of “Autonomous Everything” throughout the 2022 calendar year of OODA research and analysis:
“Handing over the figurative and literal keys: Every year more robots enter our businesses and private lives. One of the most visible examples of advancing automation is Tesla’s Full Self Driving (FSD) capability, which leverages advanced computer vision, machine learning, neural networks, and control of a self-contained electric vehicle. Other emerging examples include a humanoid robot called digit from Agility Robotics that is designed to work with humans in a factory or retail environment. The rise of smart cities is fueling more connections to physical systems such as automated traffic flow and water and power management. AI and machine learning are also automating data analytics and improving IT-based workflows and we expect improved support for automated workflows by complex physical and tech systems.”
This post expands on this initial position and our ongoing research on the future of autonomous systems.
Background
George Leopold, technology writer of the EETimes, best captured the current combination of the “trough of disillusionment” and “hype cycle” surrounding the future of autonomous systems:
“Engineering reality is catching up with the promise—much of it, for now, empty—of autonomous technologies. At what point autonomous machines can be safely let off their leash by human operators remains a nettlesome technical, legal, and, ultimately, ethical question.
Despite the claims of a growing number of ambitious technology startups, some backed by electronics industry giants, true autonomy remains elusive, reserved mostly for mundane tasks. Safe autonomous systems deployed in mission-critical applications like self-driving vehicles—that is, no room for error, no single point of failure—remain a long way off.
As autonomy hits the streets, factory floor, and, increasingly, the battlefield, regulators, and policymakers are scrambling to keep up. As with risky, high-reward domains like human spaceflight, the stress test for autonomy developers, their business models, and regulators struggling to keep pace with development and deployment will be when—not if—someone is killed by an autonomous system.” (1)
The fact remains that a Version 1 operating system for autonomous systems will need to emerge on which to build an ecosystem of “autonomous everything”. Ideally, the Autonomous Systems Operating System (ASOS) of the future (and the “OSI’ layers and stack on which it is built) will need to be not low latency, but zero latency with no-fault error data throughputs (which will in some early versions need to provide software-based predictive data corrections through ‘last mile’ data analytics, AI architectures, and data-trained models ) – all with a satellite-based and 5G-plus cellular encrypted hardware layer on which to run it.
Marilyn Wolf, an engineering professor and director of the School of Computing at the University of Nebraska–Lincoln, provides a more technical description of the problem set: “Methodologies like ISO 26262 (an automotive functional safety standard based on the V process) wrap more general assurance methods around analytic methods like control theory that characterize specific cases such as step response. These methodologies rely on characteristics like continuity allowing inference of system behavior in parts of the design space that haven’t been directly analyzed. Modern ML systems don’t have those characteristics—a very small change in input can result in a completely different output.” (1)
To use an analogy from the build-out of the “seamless streaming” video distribution architecture of the last twenty years: We cannot deploy Version 1.0 of an autonomous system operating system which resembles the 320×200 video quality (and delivery quality lack thereof) of early internet video (read: Real Networks) – or even some of the ongoing delivery issues of any of the multichannel video programming distributors (MVPDs) or online video distributors (OVDs). Oh, and for the ASOS of the future, software and hardware agnostic interoperability are table stakes: Hulu will need to “talk” to the Netflix, will need to “talk” to the DirectTV of this autonomous systems architecture.
William Widen, a professor at the University of Miami School of Law, provides further context: “I am worried about what disclosure is appropriate for purchasers of an autonomous vehicle (AV). If part of the functionality of the AV is ‘rule based’ as reflected in an algorithm, I think I have a better idea of how I might accurately describe that aspect of the AV as disclosure for consumers…. The AV industry already shows signs of an inability or unwillingness to expressly identify standards for deployment.” (1)
Waiting patiently for the equivalent of a video to haltingly load to your local RAM or waiting for the cable guy for autonomous systems repairs translates into the nanosecond decision-making process of an autonomous vehicle careening through a crosswalk during a red light or an autonomous drone delivery system falling from the sky and wreaking havoc. And there will not be a product development cycle that provides software development update windows for the entire population to alpha and beta test the product (again, think internet video or the iterative data created by the worldwide distribution of vaccinations for Covid-19 on which to develop vaccine updates. For autonomous systems, the risk profile is completely a completely different timeframe for the end-user.
Consider Benjamin Kuipers, an engineering and computer science professor at the University of Michigan, perspective on the issue: “One perspective on AI and robotics technology, including autonomous vehicles (AVs), treats them just like other potentially impactful technologies such as nuclear power and genetic engineering. We all expect problems with assessing the true safety of AVs as the time approaches to decide whether to deploy AVs at scale. Developing standards and metrics will be a central problem for the AV industry.” (1)
The burden of the “autonomous vehicle fatal accidents as feedback loop” cannot be distributed out to the consumer during an iterative production cycle. Small scale success (short and long-haul trucking, for example, will have to be scaled up once the fatal risk is close to 100% mitigated. Or will the market sustain a 95% to 99% accuracy rate, with the unintended consequences and the inevitable autonomous systems accidents that most assuredly entail?
Phil Koopman, an assistant professor of electrical and computer engineering at Carnegie Mellon University, nails down this issue further: “A significant issue with this technology is the unknowns. If you’re fundamentally taking an ML approach of training on things you have seen, what happens when you inevitably encounter one of those famous unknown unknowns that you didn’t see in training or testing? Worse, what if we find out that the number of unknown unknowns is itself unknowable?” (1)
What Next? The Advanced Autonomous Systems Hardware Layer
Case Study: Xona Space Systems – Xona Passes Testing Milestone for Navigation Demo Mission
As reported by Ryan Duffy for @Payloadspace:
Getting around: The global navigation satellite systems (GNSS) in operation right now are owned and operated by governments. The US has GPS, the EU has Galileo, and Russia and China each operate their own positioning systems. These satellites reside in a medium Earth orbit at ~20,000km and, in GPS’ case, were designed half a century ago.
These positioning, navigation and timing (PNT) constellations are integral to how we get around down on Earth. But new technologies, including autonomous vehicles, call for a higher-precision navigation system with higher security.
Enter Xona: The California-based company aims to be the first commercially operated GNSS constellation from the US. Its planned Pulsar constellation of next-generation navigation satellites will live in LEO, about 20x closer to the ground than legacy GNSS.
“There are all sorts of challenges, especially when you’re trying to get down to centimeter-level precision,” Brian Manning, Xona CEO, told Payload. “First of all, you have to know exactly where your satellites are to that level of precision. And then you have to be able to communicate that down to earth, also without losing any precision. That hasn’t really ever been done before from low Earth orbit.”
The move to LEO, while it will allow higher precision and data flow, introduces new challenges with accurate timing. “Navigation is all about timing,” Manning said. “Everything has to be time-synchronized to a nanosecond level.”
- GPS and other GNSS constellations are equipped with heavy, bulky clocks on board to keep them in step, Manning said.
- Xona leverages a patented atomic clock technology to keep its systems in sync and to offload the bulk of physical clocks to the cloud.
Hands-off driving: Xona’s satellites are designed with the requirements of autonomous driving in mind. “Our company mission is really formed around enabling modern tech to operate safely,” Manning said. “Safety is kind of a sliding scale, and it’s a tough thing to really pin down as to how safe is safe enough.”
The company’s CTO, Tyler Reid, comes from Ford’s autonomous driving unit and had a significant role in working out the standards for what kind of positioning precision is needed to support safe autonomous driving. (Talk about recruiting from the source.) Turns out, to match or exceed the safety standards for other forms of mass transit, self-driving cars need to collide less than once per one billion miles for the navigation system alone.
- Since the navigation component isn’t the only fallible element at play in a self-driving car, the vulnerability of each individual system has to be particularly low in case any one system fails.
- For example, if the vision system built into your autonomous vehicle fails, you’ll still have redundancy if your PNT system can pinpoint your location and speed to a very high level of accuracy.
“We need centimeter precision for the age of autonomy,” said Jeff Crusey, a former Seraphim investor who spearheaded the fund’s investment in Xona before leaving last month to start his own firm. Xona, Crusey says, has struck on “highly differentiated tech that is very hard to replicate.”
Huginn news: The company raised $8M in September from Seraphim Space Investment Trust and MaC Venture Capital to fund its first mission, Huginn. Xona has passed off that first mission and is hands-off until commissioning begins soon after launch when they’ll test all the onboard systems and are ready to begin the demo.
- “We broke all sorts of stuff” during the testing phase, Manning said. “We broke hardware, we broke software, we broke electronics—anything you can think of, we probably broke it. But that’s also how you go so fast. Fail fast and fix what failed.”
Coming up: Eventually, Xona aims to have a constellation of ~300 PNT satellites in LEO, but Manning says they can begin services with about 40 sats in orbit, which would give coverage to over ~70% of the world’s population. (2)
Related Reading:
Explore OODA Research and Analysis
Use OODA Loop to improve your decision-making in any competitive endeavor. Explore OODA Loop
Decision Intelligence
The greatest determinant of your success will be the quality of your decisions. We examine frameworks for understanding and reducing risk while enabling opportunities. Topics include Black Swans, Gray Rhinos, Foresight, Strategy, Stratigames, Business Intelligence, and Intelligent Enterprises. Leadership in the modern age is also a key topic in this domain. Explore Decision Intelligence
Disruptive/Exponential Technology
We track the rapidly changing world of technology with a focus on what leaders need to know to improve decision-making. The future of tech is being created now and we provide insights that enable optimized action based on the future of tech. We provide deep insights into Artificial Intelligence, Machine Learning, Cloud Computing, Quantum Computing, Security Technology, Space Technology. Explore Disruptive/Exponential Tech
Security and Resiliency
Security and resiliency topics include geopolitical and cyber risk, cyber conflict, cyber diplomacy, cybersecurity, nation-state conflict, non-nation state conflict, global health, international crime, supply chain, and terrorism. Explore Security and Resiliency
Community
The OODA community includes a broad group of decision-makers, analysts, entrepreneurs, government leaders, and tech creators. Interact with and learn from your peers via online monthly meetings, OODA Salons, the OODAcast, in-person conferences, and an online forum. For the most sensitive discussions interact with executive leaders via a closed Wickr channel. The community also has access to a member-only video library. Explore The OODA Community
